Date: Mon, 25 Feb 2008 21:17:56 +0100 From: Kris Kennaway <kris@FreeBSD.org> To: Oliver Herold <oliver@akephalos.de>, freebsd-questions@freebsd.org, "freebsd-performance@freebsd.org" <freebsd-performance@freebsd.org> Subject: Re: FreeBSD bind performance in FreeBSD 7 Message-ID: <47C32274.2060706@FreeBSD.org> In-Reply-To: <20080225133347.GA2446@asgard.home> References: <20080225133347.GA2446@asgard.home>
next in thread | previous in thread | raw e-mail | index | archive | help
Oliver Herold wrote: > Hi, > > I saw this bind benchmarks just some minutes ago, > > http://new.isc.org/proj/dnsperf/OStest.html > > is this true for FreeBSD 7 (current state: RELENG_7/7.0R) too? Or is > this something verified only for the state of development back in August > 2007? I have been trying to replicate this. ISC have kindly given me access to their test data but I am seeing Linux performing much slower than FreeBSD with the same ISC workload. http://people.freebsd.org/~kris/scaling/bind-pt.png Summary: * FreeBSD 7.0-R with 4BSD scheduler has close to ideal scaling on this test. * The drop above 6 threads is due to limitations within BIND. * Linux 2.6.24 has about 35% lower performance than FreeBSD, which is significantly at variance with the ISC results. It also doesn't scale above 3 CPUs. * I am trying to understand what is different about the ISC configuration but have not yet found the cause. They were testing 2.6.20.7 so it is possible that there was a major regression before the 2.6.22 and .24 kernels I tested. Or maybe something is broken with the Intel gige driver in Linux (they were using broadcom hardware). The graph is showing performance over 10ge, but I get the same peak performance over gige when I query from 2 clients (the client benchmark is very sensitive to network latency so a single client is not enough to saturate BIND over gige). * 7.0 with ULE has a bug on this workload (actually to do with workloads involving high interrupt rates). It is fixed in 8.0. * Changes we have in progress to improve UDP performance do not help much with this particular workload (only about 5%), but with more scalable applications we see 30-40% improvement. e.g. NSD (ports/dns/nsd) is a much faster and more scalable DNS server than BIND (because it is better optimized for the smaller set of features it supports). Kris
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?47C32274.2060706>